1

Earned-Secure Attachment Status in Retrospect and Prospect

Glenn I. Roisman

Elena Padrón

L. Alan Sroufe

Byron Egeland

In press, Child Development

to appear July/August, 2002

PLEASE DO NOT QUOTE OR DISTRIBUTE WITHOUT CONSENT OF AUTHORS

2

Abstract

Past research with the Berkeley Adult Attachment Interview demonstrates that

retrospectively defined earned-secures (who coherently describe negative childhood experiences)

parent as effectively as do continuous-secures (who coherently describe positive childhood

experiences) but manifest liabilities in the form of depressive symptomatology. This paper

presents data from a 23-year longitudinal study that replicate and extend prior research, testing a

key premise that earned-secures so defined actually have a history of insecure attachments that

change over time and/or endure consistently harsh or ineffective parenting in their youth.

Discrepant with assumptions, retrospective earned-secures were not more likely than continuous-

secures to have been anxiously attached in infancy and were observed in childhood and

adolescence to have encountered among the most supportive and structured maternal parenting in

a high-risk sample. Prospectively defined earned-secures (operationalized using participants’

infant attachment classifications) did indeed go on to have success in their close relationships,

many without reporting relatively high levels of internalizing distress in adulthood.

3

Earned-Secure Attachment Status in Retrospect and Prospect

“I had a weak father, domineering mother, contemptuous teachers, sadistic sergeants, destructive

male friendships, emasculating girlfriends, a wonderful wife, and three terrific children.

Where did I go right?” – Jules Feiffer, illustrator and satirist

Although it is becoming increasingly clear that the quality of early parent-child

attachments can manifest significant and substantial continuity over time (Hamilton, 2000;

Waters, Merrick, Treboux, Crowell, & Albersheim, 2000), empirical evidence suggests also that

change is not only possible, but, as Bowlby (1973) envisioned, meaningfully associated with

variations in the presence and timing of positive and negative life events (Weinfield, Sroufe, &

Egeland, 2000; Thompson, 2000). Of particular interest in the study of change in attachment

security are developmental processes by which individuals rise above malevolent parenting

histories to break the intergenerational cycle, typically referred to as “earned-security” (Pearson,

Cohn, Cowan, & Cowan, 1994; Phelps, Belsky, & Crnic, 1997; Paley, Cox, Burchinal, & Payne,

1999; Sroufe, Carlson, Levy, & Egeland, 1999). Despite the obvious importance of

understanding such resilient pathways, however, empirically studying positive changes in

individuals’ attachments is quite a methodological challenge, often requiring tentatively

accepting speculative findings from cross-sectional research. Few would disagree that,

ultimately, prospective data must be brought to bear on the key inferences of such studies,

especially those suppositions relevant to the validity of the retrospective assessment of early

adversity. As such, this paper presents data drawn from a 23-year longitudinal study designed to

replicate and extend pathfinding cross-sectional research on the topic of earned-security (Pearson

et al., 1994; Phelps et al., 1997; Paley et al., 1999).

4

To date, with the exception of a single set of studies described later (Waters, Hamilton, &

Weinfield, 2000), all published operationalizations of earned-security have been retrospective,

based on variations in the inferred experience ratings of those classified as secure/autonomous in

the Berkeley Adult Attachment Interview (AAI; George, Kaplan, & Main, 1985). A widely used

and well validated instrument in developmental research (Hesse, 1999), trained coders use the

AAI to make inferences about adults’ current state of mind regarding earlier attachment

experiences with their parents. Coders also typically rate individuals’ inferred childhood

experiences with five scales of parental behaviors including mother and father’s love, rejection,

neglect, role reversal, and pressure to achieve. A growing corpus of studies has now

demonstrated that adult attachment states of mind can be predicted by earlier parent-child

experiences (e.g., Allen & Hauser, 1996; Beckwith, Cohen, & Hamilton, 1999; Roisman,

Madsen, Hennighausen, Sroufe, & Collins, 2001) and in turn forecast the quality of individuals’

interactions with their own children (van IJzendoorn, 1995a). In contrast, very little is known

about the psychometric and retrodictive properties of the inferred experience scales, as it is often

assumed that they only bear a limited connection with the “actual” events of childhood (see

Pearson, Cohn, Cowan, & Cowan, 1994, for discussion).

In comparison to the “insecure” classifications (preoccupied, dismissing, unresolved),

individuals classified as secure/autonomous in the AAI produce coherent accounts of their early

experience (Main & Goldwyn, 1998). While many young adults classified as secure/autonomous

report about early supportive encounters with parents, some describe negative experiences but

nonetheless do so in a coherent and contained manner. Based in part on the observations made

by Main and Goldwyn (1998) in their coding manual for the AAI, Pearson, Cohn, Cowan, and

Cowan (1994) dubbed this latter group “earned-secure,” an implicit assumption being that they

5

had actually overcome malevolent childhood experiences— either insecure attachments in

infancy that changed over time or harsh parenting that might be otherwise associated with

insecurity in childhood and/or adolescence1.

Following a modest sized sample of mothers and fathers, Pearson and her colleagues

(1994) concluded that earned-secures so defined parent as effectively as do continuous-secures

but also manifest liabilities in the form of depressive symptomatology, presumably as a function

of their early adverse (e.g., insecure) histories. The parenting outcomes in Pearson et al.’s (1994)

study were observational in nature, tapping parents’ warmth and structure in interaction with

their preschoolers. Depressive symptomatology was assessed via self-report with the Center for

Epidemiological Studies Depression Scale (CES-D; Radloff, 1977). Importantly, in a follow-up

report on earned-security using the Pearson retrospective system of classification, earned-secures

in an all female sample were observed to parent as proficiently as did continuous-secures under

conditions of high life stress (Phelps, Belsky, & Crnic, 1997). This latter study was important in

suggesting key evidence that earned-secures had actually broken the intergenerational cycle in

that low life stress was not masking or buffering latent vulnerabilities.

As aforementioned, however, no data have yet been published to substantiate the

tentative claim that retrospectively defined earned-secures actually encounter greater adversity

than continuous-secures in infancy and/or childhood (but see Bahadur, 1998). Indeed, there is

good reason to be cautious about retrospective reports— recollection is clearly a reconstructive

process that can be strongly influenced and even distorted by subsequent experiences (Yarrow,

Campbell, & Burton, 1970; Henry, Moffitt, Caspi, Langley, & Silva, 1994). Moreover, much

research with the AAI is built upon the specific premise that the events described in the interview

are not necessarily congruent with what would have been observed by eyewitnesses. Rather,

6

individuals’ current representations of past experiences, as inferred by trained coders who

primarily assess the quality (not the content) of discourse produced about childhood memories,

are generally viewed by attachment researchers as more accurate indicators of participants’

abstracted developmental histories and thus in turn more reliably predict future caregiving (see

especially Fox, 1995, and van IJzendoorn, 1995b). Simply stated, the AAI was not designed as a

self-report measure of encounters with early adversity or support, a point emphasized in the

coding manual for the AAI when Main and Goldwyn (1998) state that the semantic content of

the AAI should not be taken as face valid as it is “not presumed that these retrospective

interviews can provide a veridical picture of early experiences” (p. 5).

To be sure, Phelps, Belsky, & Crnic (1997) were clearly concerned that, in addition to

genuine reports of early harsh experience, the AAI’s inferred experience scales might pick up

potential depression-related biases toward selectively remembering negative childhood memories

as well. This observation is especially concerning because, as discussed earlier, an empirical link

has already been demonstrated between earned-security and self-reported depressive

symptomatology (Pearson et al., 1994). Given these considerations, Phelps and colleagues

cautioned readers “the field still awaits the results of a longitudinal study that can determine

whether subjects’ actual experiences match their probable experience ratings gathered with the

AAI” (Phelps, Belsky, & Crnic, 1997, p. 34). Pearson and her co-authors (1994) were also

careful to qualify their claims, emphasizing the retrospective nature of their “diagnosis” of early

experience.

Taking these important leads from earlier research, this study presents the first

prospective, longitudinal data on the developmental antecedents and consequences of

retrospectively assessed earned- and continuous-security. We begin by attempting first to

7

replicate Pearson and colleagues’ (1994) findings demonstrating that retrospective earned-

security is associated with depression-related distress in adulthood (age 23). We next extend

these analyses backward in time in kindergarten, 1st grade, and again at age 16 via maternal and

self-report to assess how early in the life course such expected group differences might develop

and persevere. In addition, we also examine our participants’ mothers’ reports of their own

depressive symptoms, as it is possible that the depressive symptomatology reported by

retrospective earned-secures is partly a function of the across-generation transmission of

negative affect. Indeed, a family history of depressive symptomatology could be viewed as

forecasting the negative life events and/or depressogenic biases that may account for

retrospective earned-secures reporting poorer experiences with parents in the context of the AAI.

All of this remains to be clarified.

As presently small sample sizes preclude examining retrospective earned-secures’

parenting behaviors, we instead next compare retrospectively defined earned-secures,

continuous-secures, and insecures in terms of the observed quality of their romantic relationships

in young adulthood, a theoretically analogous domain of relational adaptation. Recent work

adopting an organizational perspective on human development (Sroufe & Waters, 1977; Sroufe

& Fleeson, 1986; Collins & Sroufe, 1999) strongly suggests that security is associated with

higher quality romantic relationships in adulthood (Cohn, Silver, Cowan, Cowan, & Pearson,

1992; Roisman, Madsen, Hennighausen, Sroufe, & Collins, 2001). Furthermore, one recent study

has specifically examined the quality of retrospective earned-secures’ adult partnerships, finding

that both continuous- and earned-secure wives were able to better regulate their affect in

interactions with their partners than participants who produced either preoccupied or dismissing

8

discourse regarding their early childhood experiences in the AAI (Paley, Cox, Burchinal, &

Payne, 1999).

After replicating and extending prior work, we go on to test the key premise of the

Pearson et al. (1994) retrospective system that earned-secures so defined actually have a history

of insecure attachments in infancy that change over time and/or endure harsh or ineffective

parenting in their youth. Specifically, we begin by examining whether retrospectively defined

earned-secures in a high-risk sample are more likely than continuous-secures to have had been

independently assessed as anxiously attached in the Ainsworth and Wittig (1969) Strange

Situation at 12 and 18 months. We next compare the security groups on observationally assessed

ratings of the maternal parenting they encountered in childhood and adolescence (24 months, 42

months, and 13 years). If retrospective earned-secures indeed overcame pervasively malevolent

parenting experiences, they should be more likely than continuous-secures to have been

insecurely attached in infancy and/or should have mothers who received significantly poorer

ratings in the task-centered observations we conducted in the participants’ earlier lives.

This study concludes with the first systematic prospective examination of how change in

attachment security from infancy to young adulthood is associated with salient issues in the years

of maturity. A recent report with the current sample shows that individuals can be classified into

prospective attachment change groups on the basis of their infant attachment histories;

individuals who were insecurely attached in infancy but nonetheless produce autonomous/secure

discourse in the AAI are classified as prospectively defined earned-secures (Weinfield, Sroufe, &

Egeland, 2000; see also Hamilton, 2000). As predicted by attachment theory (e.g., Bowlby,

1973; Sroufe, Carlson, Levy, & Egeland, 1999), our initial study found that prospective earned-

secures could be discriminated from other groups in terms of their life histories, evidencing less

9

maltreatment and higher quality family dynamics at age 13 than did continuous-insecures

(insecure infants that produced incoherent attachment-relevant discourse in young adulthood).

Using this prospective classification scheme, we present here complementary data on the

consequences of prospectively defined earned-security for adult adaptation. The extant

retrospective published reports on earned-security offer that individuals who overcome poor

childhood experiences are able to rise above their past in the domain of relationships but also

imply that early malevolent experience exacts a toll in terms of increased risk for depressive

symptomatology in adulthood. Due to questions about the way in which early negative

experience was assessed, however, we cannot be sure that individuals classified as earned-secure

in past studies actually encountered pervasive childhood adversity. This study uses a prospective

assessment of positive change in attachment security across time (by definition a valid

operationalization of earned-security) to examine by comparison the operational validity of

retrospectively assessed earned-security, both in terms of its necessary antecedents (harsh early

experiences) and its putative developmental consequences (e.g., depressive symptomatology,

success in salient adult relationships).

Method

Participants

Participants were drawn from an ongoing 23-year longitudinal study of normative and

atypical development originally consisting of 267 mothers and their first-born infants. Mothers,

many of whom were young and living in conditions of poverty, were recruited between 1975 and

1977 at Minneapolis public health clinics where they were receiving prenatal care (see Egeland

& Brunquell, 1979, for an early report). Specifically, however, it is a sub-sample of 170 of their

children who have been followed into young adulthood that form the core set of participants for

10

the current study. Data for this study were drawn from self-reports and videotaped observations

of these individuals and their mothers.

Regarding sample attrition, 212 families remained in the study by the time participants

were 24 months. Since age 2, 80.2% (n = 170) of the remaining sample has been retained. The

follow-up sample is racially diverse: 66.5% of the ongoing participants are European American,

17.6% have mixed racial backgrounds (European American, African American, Latino and/or

Native American), 10% are African American, 1.8% are Native American or Latino, and 4.1%

are unclassifiable due to missing data on their fathers’ ethnicity.

Retrospective Earned-Security Status

Procedure

At age 19, 170 participants completed the Adult Attachment Interview (AAI), a semi-

structured, semi-clinical assessment used to characterize individuals’ current state of mind with

respect to past parent-child experiences (George, Kaplan, & Main, 1985; see Weinfield, Sroufe,

& Egeland, 2000, for sample specific details). More specifically, the AAI is a 20-question,

approximately hour-long interview that requires participants to describe their early relationships

with their parents, revisit salient separation episodes, explore instances of perceived childhood

rejection, recall encounters with loss, and speculate about their future feelings and expectations

regarding raising their own child. According to established protocol, AAI’s were transcribed

verbatim and all identifying information was removed from the transcripts before they were

coded by judges trained through and reliable with the lab of Dr. Mary Main.

In preparation for making an overall judgment regarding participants’ primary adult

attachment classification, coders assessed several “state of mind” scales including the narrative

coherence of each transcript along a nine-point rating scale entitled “coherence of mind” (ρI =

11

.71, p < .001). Violations of any of Grice’s (1975) maxims of conversation (quality, quantity,

relation, and manner) in addition to the “nature of the subjects’ apparent belief systems in

comparison to the judge’s own assessment of reality” are considered in rating participants’

coherence of mind, viewed theoretically as the single best indicator of security in the AAI (Main

& Goldwyn, 1998, p. 108). Although inferred experience scales were not systematically codified

in the initial data reduction effort of the AAI transcripts, “state of mind” scales (including

derogation, idealization, involving anger, passivity, etc.) were rated by judges who took into

account participants’ self-reported childhood experiences with parents (e.g., to detect narrative

discrepancies indicative of idealization).

Ultimately, transcripts received primary attachment classifications of

‘secure/autonomous’ (F), ‘insecure/dismissing’ (Ds), ‘insecure/preoccupied’ (E), or ‘unresolved’

(U) according to the criteria outlined by Main and Goldwyn (1998) in their coding manual for

the AAI. Narratives coded as secure showed evidence of an autonomous state of mind with

respect to attachment. These participants explored their thoughts and feelings about earlier

parent-child experiences, whether described as good or ill, in an open, contained, and above all

coherent manner. Narratives coded as insecure in contrast provided strong evidence of

dismissing, preoccupied, or unresolved states of mind with respect to attachment. These

participants idealized/minimized attachment relationships, seemed currently entangled/enmeshed

in their relationships with parents, or their discourse became disorganized when describing loss

or abuse events, respectively. Percent agreement for the secure/insecure designation was 87.8%

(κ = .72, p < .001). Keeping with previous work with the retrospective earned-security

classification (e.g., Phelps, Belsky, & Crnic, 1997), unresolved participants whose secondary

attachment classification was secure/autonomous (F) were dropped from further analysis as their

12

discourse could not be unambiguously described as indicative of either a secure or insecure state

of mind regarding attachment.

Earned- and Continuous-Secure Groups. After identifying secure (n = 46) and insecure (n

= 107) transcripts2, the secure/autonomous group was further sub-divided into retrospectively

defined earned- and continuous-secures by scrupulously following a procedure developed by

Pearson, Cohn, Cowan, and Cowan (1992) (see also Phelps, Belsky, & Crnic, 1997). In order to

follow this procedure, all secure transcripts as well as a sub-set of randomly chosen insecure

transcripts (n = 25) were rated on several nine-point scales designed to measure participants’

inferred childhood experiences by classification-blind coders trained through the lab of Mary

Main. These scales included mother and father’s love, rejection, neglect, and pressure to achieve.

As described in the Introduction, these ratings are meant to provide an overall depiction of

participants’ experiences with their primary caregivers in childhood. Intraclass reliabilities for

these scales (ρI’s) were adequate to excellent, ranging from .77 to .92 (p’s < .001). All ratings

were double coded for the secure cases and scores were averaged to increase reliability.

As in the Pearson et al. (1994) study, an initial attempt was made to define earned-

security by examining secure/autonomous transcripts with the following sub-classifications: F1,

F2, F3b, F4, and F5, codes assigned on the basis of high narrative coherence in combination with

a negative inferred developmental history (Main & Goldwyn, 1998). Upon careful inspection of

the childhood experience scores of these transcripts, however, it was discovered that some of

these participants described only somewhat difficult childhoods that did not suggest the same

degree of hardship encountered by insecures and others designated as earned-secure. Transcripts

containing only weak evidence for difficult early experiences were coded as continuous-secure.

13

As in previous research on earned security (e.g., Pearson et al., 1994; Phelps et al., 1997),

earned-secures were thus ultimately defined as participants who produced coherent (e.g., secure)

discourse during the AAI but whose mother and/or father received low scores on the loving scale

and whose mother and/or father received a high rating either on the rejecting or neglecting

inferred experience scales. For the purposes of this study, ratings below the midpoint (‘5’) on the

inferred childhood experience scales were defined as “low” and ratings above the midpoint on

the scales were considered “high.” The final group of earned-secures (n = 24) consisted of

participants with the following F codes: F1, F2, F3b, F4, F5, and F other. Continuous-secures (n

= 22) had the following sub-classifications: F2, F3a, and F4 (primarily F4a). Note that one

transcript was designated earned-secure due to low loving scores combined with marked role-

reversal on the part of the participant’s mother (as well as sub-threshold rejection and neglect). In

two cases, the profile of ratings between coders resulted in discrepancies about whether the case

should be regarded as earned- or continuous-secure. As such, a third coder rated these cases and

group designation (earned- or continuous-secure) was decided on the basis of agreement between

two of the three coders.

Before moving forward with analyses, a set of planned comparisons was conducted to

examine empirically whether earned-secures had significantly poorer childhood experience

ratings than continuous-secures and a significantly higher coherence rating than insecures as is

required by the design of this study. Consistent with previous research using the Pearson et al.

(1994) method of retrospectively operationalizing earned- versus continuous-security, results of

planned comparisons provided robust evidence that the Pearson system appropriately

differentiated the three security groups with respect to the inferred early experience and narrative

coherence ratings. Specifically, t-tests revealed that retrospectively defined earned-secures had

14

significantly poorer experience ratings than continuous-secures (see Table 1; t [38] = 2.39, p <

.05 for mother loving; t [44] = 8.38, p < .001 for father loving; t [28] = 2.66, p < .05 for mother

rejecting; t [33] = 6.29, p < .001 for father rejecting; t [24] = 3.88, p < .001 for mother

neglecting; and t [44] = 8.90, p < .001 for father neglecting) and were scored as significantly

more coherent than insecures (t [31] = 11.56, p < .001)3. Note that, with the exception of the

father loving and neglecting analyses, the statistics reported here do not assume equal variances

across groups as Levene’s test of homogeneity of variances was significant for these analyses.

In addition, data from the randomly chosen sub-sample of insecure transcripts coded for

inferred childhood experiences (n = 25) revealed that their inferred experience means, as

expected, were similar to those of the earned-secures (see Table 1; t [47] = 1.75, p = .09 for

mother loving; t [44] = 2.19, p < .05 for father loving; t [42] = .95, p = .35 for mother rejecting; t

[38] = 4.68, p < .001 for father rejecting; t [44] = 1.68, p = .10 for mother neglecting; and t [31]

= 4.23, p < .001 for father neglecting)4. In fact, the only significant differences observed between

these two groups revealed that earned-secures indicated significantly greater rejection and

neglect as well as less love by their fathers than did insecures (a finding with precedent in Paley

et al., 1999). Only one analysis intimated that earned-secures’ experience ratings might not be

comparable to or worse than insecures: Echoing data presented in Phelps et al. (1997) and Paley

at al. (1999), earned-secures had a higher group mean on mother loving than did insecures,

although this difference was marginally significant. It should be emphasized as a counterpoint,

however, that earned-secures were also rated as experiencing marginally more neglect from

mothers than did insecures.

______________________________________________________________________________

Insert table 1 about here

15

______________________________________________________________________________

In addition to the Pearson et al. (1994) method of operationalizing earned-security, an

attempt was made to sort participants into retrospectively defined earned- and continuous-

security groups by following the more conservative criteria advocated for by Main and Goldwyn

(1998) in the most recent edition of the AAI coding manual. Specifically, earned-secures are now

defined in the AAI manual as secure/autonomous participants who receive loving scores lower

than 2.5 for both mother and father figures and continuous-secures are secure/autonomous

participants whose parents’ average loving scores are 6.5 or greater. Main and Goldwyn suggest

that participants classified as autonomous who fail to meet either of these criteria should be set

aside from analyses.

A quick look at the inferred experience ratings for this cohort revealed that only three

participants met Main and Goldwyn’s (1998) stringent standard for retrospectively

operationalizing earned-security (ten participants met criteria for continuous-security). As it

would be statistically inappropriate to report a set of analyses complementary to those presented

for the Pearson et al. (1994) system given the small earned-secure group n, no analyses using the

Main and Goldwyn (1998) retrospective classification criteria are reported here.

Prospective Earned-Security Status

Prospectively defined earned- and continuous-secure groups were created on the basis of

the observed infant attachment histories of participants who produced autonomous discourse in

the AAI. Specifically, at 12 and 18 months, participants and their mothers completed the well-

validated Strange Situation (SS) behavioral assessment of attachment security (Ainsworth,

Blehar, Waters, & Wall, 1979). At both ages, mother-child dyads received one of three primary

16

attachment classifications (secure, avoidant, or resistant) based on the infants’ responses during

separation and reunion from their primary caregivers. Dyads were coded as ‘secure’ when the

infant used the parent as a secure base from which to explore— upon reunion, these infants’

interactions with their caregivers served to alleviate separation distress. In non-secure dyads, the

infants did not use parents as a secure base— infants either avoided the caregiver upon reunion

(‘avoidant’) or displayed distress that was not effectively alleviated (‘resistant’). For the

purposes of this study, all non-secure dyads were coded as insecure.

Following precedent established in a previous study with the current data set (Weinfield,

Sroufe, & Egeland, 2000), two groups were created by cross-classifying attachment security

status in infancy and adulthood: continuous-secure (secure in SS-secure in AAI) and earned-

secure (insecure in SS-secure in AAI). Two separate variables were created: security change

status from 12 months to adulthood (n = 26 continuous-secure, n = 19 earned-secure) and

security change status from 18 months to adulthood (n = 31 continuous-secure, n = 11 earned-

secure). Participants who produced insecure discourse in the AAI (irrespective of their infant

attachment history) formed the comparison group for analyses (n = 107; 58 secure in 12-month

SS, 55 secure in 18-month SS).

Earned-Security: Developmental Antecedents

Overview

All variables described below are grouped into theoretically relevant categories key to

examining the operational validity of the retrospective earned-security system developed by

Pearson et al. (1994). Measures of internalizing distress are discussed first. Next we describe the

three observational assessments of maternal parenting conducted with our participants in early

childhood and adolescence.

17

Depressive Symptomatology

Child Behavior Checklist/Youth Self-Report. Mothers of our participants completed the

commonly used and well-validated Child Behavior Checklist (CBCL) to describe their children’s

problem behaviors, adaptive functioning, and school performance when participants were in

kindergarten, grade 1, and again at age 16 (see Achenbach, 1991a, 1991b for details). At age 16,

participants themselves completed the parallel form of the CBCL known as the Youth Self-

Report (YSR). Both the CBCL and YSR were normed on a large, representative national sample

of non-referred children and have adequate psychometric properties including excellent test-

retest reliability, discriminant validity for the sub-scales, and criterion-related validity (e.g., the

measures distinguish between referred and non-referred children). For the present study, the

internalizing factor, containing items that tap behaviors relating to anxiety/depression, social

withdrawal, and somatic complaints, were used in analyses. Raw scores were converted to

standardized T-scores: Mother-report data from kindergarten and 1st grade were composited to

create a single variable for analyses; age 16 CBCL and YSR internalizing scores were examined

separately by informant.

CES-D. Mothers were asked questions regarding their own depressive symptomatology

when their children were 42 months with the 20-item Center for Epidemiological Studies

Depression Scale (CES-D; Radloff, 1977). The CES-D has acceptable psychometric properties

including adequate convergent validity (e.g., with the Bradburn Negative Affect Scales;

Bradburn, 1969) and criterion validity (e.g., it discriminates between the general population and

18

inpatient populations). CES-D scores of 16 and above are considered clinically significant

(Radloff, 1977).

Observed Quality of Parenting Experiences: Childhood and Adolescence

Maternal support and structure at age 2. At 24 months, participants and their mothers

completed a set of “tool tasks” adapted from the ethological research of Bill Charlesworth (see

Matas, Arend, & Sroufe, 1978). These tasks, requiring the 2-year-olds to complete four problem-

solving exercises, increased in complexity, ultimately becoming too difficult for the child to

complete on his/her own. Assessments were videotaped and subsequently coded for mother’s

supportive presence, quality of assistance, and limit setting during a cleanup task that preceded

the tool problems. These assessments were independently coded by two graduate research

assistants yielding an average interrater r = .87. As in a prior study with the current dataset

(Aguilar, Sroufe, Egeland, & Carlson, 2000), the quality of mothers’ provision of support and

structure was quantified by standardizing and averaging two variables: “limit setting” during

cleanup and the “overall mother rating” on the tool tasks (see Appendix).

Maternal support and structure in preschool. At 42 months, participants and their mothers

completed a set of four “teaching tasks” adapted from a procedure developed by Block and

Block (1980). As in the tool tasks, the teaching task exercises required the joint attention of

mothers and their preschoolers for their successful completion. Specifically, problem solving

tasks consisted of block design, object naming, matrix sorting, and Etch-a-Sketch drawing.

Paralleling the 24-month tool task, this assessment was videotaped and a set of mother behavior

ratings was scored by two independent raters (the average interrater Pearson correlation was .76).

In accord with a previous compositing strategy used by this research project (Aguilar et al.,

19

2000), four of the mother behavior scales (respect for autonomy + structuring and limit setting +

supportive presence - hostility) were composited and standardized to create a preschool index of

maternal support and structure (α = .83; see Appendix).

Quality of mother-adolescent relationship. At age 13, participants completed a videotaped

protocol with their primary caregiver. Tasks involved creating an anti-smoking campaign,

completing a puzzle with the caregiver blindfolded, discussing results of imaginary events, and

collaborating on a Q-Sort of the ideal person. Graduate research assistants coded videotapes of

the parent-child interactions using eleven dyadic rating scales of behavior and affect (J. Sroufe,

1991). These rating scales included Anger, Conflict, Conflict Resolution, Confrontive-Attacking,

Emotional Engagement, Hostility, Negative Affect, Positive Affect, and three “Balance” scales

that that focused on the degree to which relationships served to promote the expression of

individual ideas, scaffold personal development, as well as help individuals meet task demands.

Intraclass correlations (ρI’s) were adequate, ranging from .60 to .70 (p’s < .001). Several

additional rating scales reported elsewhere (cf. Aguilar et al., 2000) rating “generational

boundary violations” were excluded from analysis here as they are not regarded by the project as

mother-child relationship quality measures per se.

A previous study with the current sample has shown that security in the AAI is associated

with higher quality parent-adolescent interactions in participants’ earlier lives, assessed using a

composite measure of “mother-child process,” derived from a principal components analysis of

the rating scales mentioned above (see Roisman, Madsen, Hennighausen, Sroufe, & Collins,

2001). Specifically, the composite rating is a standardized average of the Balance I, Balance III,

Emotional Engagement, and Positive Affect scales (α = .83, see Appendix). Mother-child

process was used as a measure of the quality of mother-child relationships at age 13 in order to

20

explore retrospective earned-secures’ parenting experiences in comparison to continuous-secures

and insecures.

Earned-Security: Young Adult Sequelae

Overview

Previous research suggests that retrospectively defined earned-security is associated with

high quality relationships (Pearson et al., 1994; Phelps et al., 1997; Paley et al., 1999) and

depressive symptomatology (Pearson et al., 1994) in adulthood. The measures described below

were used to replicate and extend prior research within the framework of a longitudinal study.

Internalizing Distress

Young Adult Self-Report. At age 23 as part of a larger follow-up assessment of

adaptation in young adulthood, participants completed the Young Adult Self-Report (YASR), the

adult form of the Child Behavior Checklist (Achenbach, 1997). Analyses for this study were

focused on a sub-set of the YASR probes that load on an internalizing factor, containing items

related to both depression/anxiety and social withdrawal. Raw scores were converted to

standardized T-scores for analyses. Note that T-scores of 60 and above are considered clinically

significant (Achenbach, 1997).

Observed Quality of Romantic Relationship

Romantic relationship assessment. At age 20-21, participants were invited to complete a

romantic relationship assessment with their partners of four months or longer. Tasks included

discussing a couple-identified problem and collaborating on an ideal couple Q-sort. Seventy-

three participants (36 male participants with their girlfriends and 37 female participants with

21

their boyfriends) completed the entire assessment, 61 of whom had earlier completed the AAI at

age 19. Note that the demographics of the participants in this sub-sample are comparable to those

of the full follow-up longitudinal sample (Hennighausen, 1999).

Graduate research assistants coded videotapes of the couple interactions using ten dyadic

rating scales of behavior and affect developed to parallel parent-child ratings used previously by

the project when participants were age 13 (Aguilar, Christian, Collins, Cook, Hennighausen,

Hyson, Levy, Meyer, Roisman, Ruh, Sesma, & Vogeler-Knopp, 1997). These scales included

Anger, Conflict Resolution, Dyadic Negative Affect, Hostility, Overall Quality, Secure Base,

Shared Positive Affect, and three “Balance” scales that paralleled conceptually those used in the

age 13 mother-child assessment described earlier (ρI’s ranged from .81 to .95, p’s < .001).

Coders were blind to the identity of the original participants within the couples.

Romantic relationship quality has been previously operationalized in this research project

using a composite measure of “romantic relationship process” derived from a principal

components analysis of the rating scales mentioned above (Roisman, Madsen, Hennighausen,

Sroufe, & Collins, 2001). This quality measure is a standardized average of the Balance I,

Balance II, Conflict Resolution, Overall Quality, Secure Base, and Shared Positive Affect rating

scales (α = .95, see Appendix). Previously reported analyses have shown that participants coded

as secure at age 19 were involved in higher quality romantic relationships at age 20-21 than

insecures (Roisman et al., 2001). The romantic relationship process composite variable is used in

the current study to determine whether retrospective earned-secures are involved in adult

romantic relationships of comparable quality to continuous-secures and of higher quality than

insecures, a finding expected on the basis of previous research (e.g., Paley et al., 1999).

Results

22

Analytic Plan

With the exception of cross-tabulations examining whether retrospectively and

prospectively defined earned- and continuous-secure groups were equivalent, all analyses

described below are planned comparisons. Employing the analytic template of prior research on

earned-security (e.g., Pearson et al., 1994), planned comparisons were used here as they are ideal

in protecting against Type II error (e.g., they eliminate the need for omnibus F-tests). Although

ANOVAs with post hoc follow-ups are to be preferred in exploratory research (in which Type I

error rates are conservatively minimized), planned comparisons should be used when focusing

on a well defined set of a priori contrasts of interest, as was the case here (see Hays, 1988).

As a maximum of k-1 comparisons can be completed per outcome variable using planned

comparisons (where k is defined as the number of levels in the independent variable), each set of

analyses described below contrasts the earned-secures with the other two security groups (e.g.,

earned-secures versus continuous-secures, earned-secures versus insecures). Two-tailed

significance values are reported for all analyses; in addition, one-tailed p-values are indicated

when directional hypotheses were explicitly articulated a priori (e.g., for mother and child

internalizing distress and romantic relationship quality replication analyses). In addition to t-

statistics and their associated p-values, standardized differences between means (Cohen’s d;

effect sizes) were calculated and are indicated for all contrasts. We adopted Cohen’s (1992)

criteria in interpreting d (small effect = .2, medium effect = .5, and large effect = .8 +) so that

marginally significant differences could be considered in relation to their effect size in

determining the substantive significance of findings (cf. Cohen, 1994). In prior research on

earned-security (Phelps et al., 1997), d’s ≥ .50 were considered meaningful effects. Note finally

that n’s vary across analysis to maximize data available from each assessment (see Tables 2-5).

23

Prospective versus Retrospective Earned-Secure Attachment Status.

We begin by emphasizing that the prospective and retrospective assessments of earned-

security did not converge: Two crosstabulations of the retrospective operationalization of earned-

versus continuous-security by each of the two prospective operationalizations revealed that the

Pearson et al. (1994) retrospective system had both extremely low sensitivity and specificity in

accurately categorizing which participants actually had insecure attachments in infancy that

changed over time and which participants had secure attachments in infancy and again at 19

years of age (see Table 2). Specifically, kappas for both analyses were negative, though non-

significant (12-month Strange Situation prospectively defined secure groups by retrospectively

defined secure groups percent agreement = 44.4%; 18-month Strange Situation prospectively

defined secure groups by retrospectively defined secure groups percent agreement = 47.6%).

______________________________________________________________________________

Insert table 2 about here

______________________________________________________________________________

Attachment classifications by sex

When earned security was operationalized prospectively on the basis of infant Strange

Situation classifications, no sex differences emerged. In contrast, a marginally significant chi-

square revealed that the distribution of retrospectively defined earned- and continuous-secures

differed as a function of participants’ sex. Inspection of the cross-tabulations revealed that

female participants were more likely than males to be coded retrospectively defined earned-

secure and males were more likely than females to be coded continuous-secure in this sample

(see Table 3).

______________________________________________________________________________

24

Insert table 3 about here

______________________________________________________________________________

Retrospectively Defined Earned-Security: Young Adult Sequelae

Observed Quality of Romantic Relationship. As expected, retrospectively defined earned-

secures were rated as involved in higher quality romantic relationships at age 20-21 than

insecures (a large effect, t [58] = 2.04, p = .07 [p < .05 one-tailed]; d = .79). Means for the

earned-secure and continuous-secure groups on romantic relationship quality were not

discriminable (t [58] = .11, p = .91; d = .05; see Table 4).

Internalizing Distress in Adulthood. Planned comparisons revealed that earned-secures

reported marginally more distress at age 23 than did continuous-secures (a medium sized,

meaningful effect; t [38] = 1.67, p = .10 [p < .10 one-tailed]; d = .51). The earned-secure and

insecure group means did not differ significantly (t [34] = .16, p = .87; d = .04. Note also that as

Levene’s test was marginally significant for this set of analyses, the t statistics reported do not

assume equal variances across groups. See Table 4 for means and standard deviations.

______________________________________________________________________________

Insert table 4 about here

______________________________________________________________________________

Retrospectively Defined Earned-Security: Developmental Antecedents

Depressive Symptomatology. Analyses presented in Table 5 reveal a predictable pattern

of higher mother- and self-reported levels of internalizing distress in childhood and adolescence

among earned-secures relative to other groups. Specifically, in kindergarten/grade 1, mothers

reported that earned-secures had the highest levels of internalizing distress of any security group;

25

contrasts revealed significant differences between retrospective earned-secures and both

insecures (t [143] = 2.09, p < .05; d = .51) and continuous-secures (t [143] = 2.14, p < .05; d =

.64). Similar though somewhat less consistent results obtained in adolescence. At age 16, while

mothers of earned-secures rated their adolescents as more distressed on average than insecures

and continuous-secures, these differences were marginally significant only when considering

one-tailed significance levels, though the effect size was medium for the earned-secure versus

continuous-secure contrast (earned-secures v. insecure: t [141] = 1.57, p = .12 [p < .10 one-

tailed]; d = .36; earned-secure v. continuous-secure: t [141] = 1.38, p = .17 [p < .10 one-tailed]; d

= .48). Likewise, retrospectively defined earned-secure participants rated themselves as the most

distressed group in adolescence, marginally more so than insecures (t [142] = 1.87, p = .06 [p <

.05 one-tailed]; d = .41) and continuous-secures (t [142] = 1.53, p = .13 [p < .10 one-tailed]; d =

.48). Effect sizes for these analyses approached medium by Cohen’s (1992) criteria. Lastly,

group comparisons of mothers’ reports of their own depressive symptomatology revealed that,

when their children were 42 months, mothers of retrospective earned-secures indicated

significantly more distress than mothers of insecures (t [140] = 2.23, p < .05; d = .49) and

continuous-secures (t [140] = 2.02, p < .05; d = .61).

Observed Quality of Parenting Experiences. Table 5 reveals that earned-secures were

consistently rated as having encountered among the highest quality maternal parenting relative to

other security groups. Specifically, at 24 months, earned-secures were observed to have received

the best support and structure of any group, significantly more so than insecures (t [122] = 2.31,

p < .05; d = .60) and marginally more so than continuous-secures (a meaningful effect, t [122] =

1.65, p = .10; d = .52). Although no group differences emerged at 42 months, the earned-secure

mean remained above the sample average (d’s indicated small effects). Finally, at age 13 earned-

26

secures had a significantly higher group mean on the mother-child relationship quality rating

than insecures (t [141] = 2.90, p < .01; d = .66). Retrospective earned- and continuous-secure

group means were not discriminable and the effect size for this contrast was small to medium (t

[141] = 1.22, p = .23; d = .38).

______________________________________________________________________________

Insert table 5 about here

______________________________________________________________________________

Prospectively Defined Earned-Security: Young Adult Sequelae

Observed Quality of Romantic Relationship. Consistent with the results of retrospective

analyses, prospectively defined earned-secures (12-month operationalization) were rated as

involved in romantic relationships of higher quality than insecures (a large effect; t [57] = 1.96, p

= .06 [p < .05 one-tailed]; d = .76) and of comparable quality to continuous-secures (t [57] = .60,

p = .55; d = .26). Note that while a similar pattern of differences emerged with respect to groups

derived from the 18-month prospective operationalization of earned-security, the number of

earned-secures in this analysis (n = 2) precluded meaningful group comparisons (see Table 4).

Internalizing Distress. No group differences emerged with respect to internalizing distress

at age 23 using either the 12- or 18-month prospective operationalization of earned-security (see

Table 4). Effect sizes were also small, ranging from .09 to .18. Moreover, post hoc follow-ups

revealed no significant differences in terms of the proportion of clinically significant cases of

distress among prospective earned-secures relative to continuous-secures or insecures.

Discussion

This study was designed with a targeted purpose: to replicate, extend, and provide added

conceptual clarity to the literature on earned-security, defined as the processes by which

individuals overcome malevolent parenting experiences. In doing so, our main success, we hope,

27

has been in demonstrating the incredible complexity of measuring, analyzing, and, perhaps most

elusive, truly understanding the nature of positive change in the quality of attachments across

time. This discussion begins by briefly reorienting readers to the principal results of the present

study and goes on to speculate about their significance for understanding resilience processes

more generally.

As predicted, this study replicated the results of groundbreaking research on earned-

security by Pearson, Cohn, Cowan, and Cowan (1994), finding that retrospectively defined

earned-secures (who coherently described negative childhood experiences) reported marginally

more internalizing distress in adulthood than did continuous-secures (who coherently described

positive childhood experiences). This paper is the first to extend such analyses backward in time

through childhood, finding a similar pattern of results since at least kindergarten and through

adolescence via maternal and self-report. Analyses of participants’ mothers’ reports of their own

depressive symptomatology when their children were 42 months old also revealed elevated

levels of distress among parents of participants who would be later retrospectively categorized as

earned-secure. Note that while sample size restrictions produced somewhat under-powered

analyses when adopting more conservative two-tailed critical t-values, effect sizes were typically

‘medium’ by Cohen’s (1992) standards and therefore encouraged an interpretation that group

differences on internalizing distress were indeed meaningful. We caution readers, however, that

effects for self- and maternal reports of adolescents’ distress were often marginally significant

both in terms of their two-tailed p-values and Cohen’s d, although group differences were

consistently in the predicted direction.

Despite manifesting depressive liabilities, but also in line with previous research on

retrospectively defined earned-security, we found that earned-secures enjoy success in their close

28

relationships in adulthood (Pearson et al., 1994; Phelps et al., 1997; Paley et al., 1999).

Specifically, this study replicated findings that earned-secures (retrospectively and prospectively

defined) have high quality adult romantic relationships; at age 20-21 earned-secures were

observed to have been engaged in romantic relationships of comparable quality to continuous-

secures and higher quality than insecures (see Paley et al., 1999). We emphasize further that the

difference between insecures and earned-secures should be considered large and meaningful as

d’s approached .8.

Having gained confidence that we had identified a group with the distinctive profile of

participants labeled earned-secure in previous cross-sectional work, we went on in the context of

this longitudinal study to examine the key issue of whether retrospectively defined earned-

secures actually encounter significantly greater adversity in childhood than continuous-secures.

To be sure, participants coded as earned-secure on the basis of Pearson, Cohn, Cowan, and

Cowan’s (1994) retrospective system report as adults having experienced greater childhood

parenting-specific adversity than other secure/autonomous participants. Nevertheless, the data

presented herein call into question the assumption that retrospectively defined earned-secures

actually encountered greater parental adversity in childhood than continuous-secures. Using the

Pearson et al. (1994) system of operationalizing earned-security, the longitudinal, prospective

data presented in this paper revealed instead that earned-secures were the beneficiaries of among

the most supportive maternal care in a high-risk sample (effect sizes were medium; for

independent replication see also Bahadur, 1998). We also emphasize to readers that, critically,

retrospectively assessed earned-secures were no more likely than continuous-secures to have had

anxious infant attachment histories: in this study, the Pearson et al. (1994) guidelines failed to

29

discriminate secure adults with insecure infant attachment histories (e.g., prospective earned-

secures) from participants with secure attachments in both infancy and adulthood.

Clearly, lacking father-child observational data on the majority of our participants (who

did not have a stable male figure in the home), we can not know with certainty whether the

supportive maternal care we observed among earned-secures took place within the context of a

more generally supportive family dynamic in childhood or if it was compensatory, leaving open

the possibility that retrospectively defined earned-secures actually do experience poorer

relationships with male figures in childhood relative to their continuous-secure counterparts. In

addition, no observations of parent-child relationships were conducted between the ages of four

and twelve, a time that is emphasized in the AAI assessment of childhood memories. As a

counterpoint, however, given the family and personal history of depressive symptomatology in

the lives of the retrospective earned-secures uncovered in this and the Pearson et al. (1994)

study, we can also not rule out the possibility that self-described differences in early experience

between retrospectively defined earned- and continuous-secures are primarily a function of

positive and/or negative reporting biases (e.g., negative attentional biases associated with

depression5).

Although the possibility remains that the retrospectively defined earned-secures of this

study actually struggled with unmeasured forms of adversity in childhood, the foregoing findings

squarely emphasize that it should not be assumed that they experience either insecure

attachments in infancy or pervasively malevolent experiences with parents in childhood. To be

sure, Pearson and her colleagues (1994) by no means presupposed that their system for

operationalizing earned- versus continuous-security would provide a veridical window on any

adult’s infant attachment history. However, it is fair to say that the Pearson system is meant to

30

provide an overall and generally applicable depiction of participants’ actual encounters with

harsh parenting in childhood. Moreover, viewed through the lens of the findings of this paper,

the fact that retrospectively defined earned-secures have been shown in prior research to parent

as effectively as do continuous-secures (and more capably than insecures) is not necessarily

remarkable— this study suggests that their parenting skill may simply be a natural extension of

the supportive care from which they benefited in their own childhood (Pearson et al., 1994;

Phelps et al., 1997). Note, however, that these findings do not call into question the validity of

the AAI. Rather, our results underscore the major premise of the instrument that it is not the

content of early memories but rather the coherence with which they are described that provides

an accurate depiction of states of mind indicative of earlier supportive and malevolent childhood

experiences.

Although not the central focus of this study, the results of prospective analyses regarding

the consequences of change in attachment security across time advocate for modifying the

prevalent view of earned-security as necessarily linked with depressive symptomatology in

adulthood (Pearson et al., 1994). When operationalized on the basis of observed change in

attachment security from infancy, this study did not find that prospectively defined earned-

secures were at any significantly greater risk for internalizing distress in adulthood relative to

other security groups. This result stands in stark contrast to research using retrospective

operationalizations of earned-security, which, if taken at face value, imply that those who

overcome insecure attachments in infancy or pervasively malevolent parenting experiences in

childhood are more likely to be distressed than other secure adults (see Pearson et al, 1994;

Phelps et al., 1997).

31

More generally, the results of this study have potentially significant implications for how

developmentalists conceptualize resilience processes (for an excellent review see Luthar,

Cicchetti, & Becker, 2000). Our view, rooted in an organizational framework, is that resilience is

an emergent property of a hierarchically organized set of protective systems that cumulatively

buffer the effects of adversity and can therefore rarely, if ever, be regarded as an intrinsic

property of individuals (Egeland, Carlson, & Sroufe, 1993). Our data are striking on this subject:

we consistently find that competent functioning in the context of adversity is a predictable

outcome of earlier successful adaptation with respect to stage salient issues whose roots lie in

earlier supportive care (cf. Egeland, Jacobvitz, & Sroufe, 1988; Sroufe, Egeland, & Kreutzer,

1990). Using the specific case illustrated in this report, for example, earned-secures (whether

retrospectively or prospectively defined) appear able to talk about early distressing events

coherently because of either consistent or ameliorative support in childhood (see also Weinfield

et al., 2000). Earned-secures do not rise above malevolent parenting through sheer will; rather,

their success is scaffolded by caring adults, their security is a natural extension of a supportive

(though not necessarily ideal) past. Said another way, individuals who overcome and

successfully integrate turbulent pasts can indeed be identified. Our thesis is simply that the

processes implicated in overcoming such adversity necessarily involve current and often long-

standing developmental supports.

Still, questions do remain regarding our finding that, at roughly the same time, mothers of

retrospectively defined earned secures were rated as highly capable caregivers yet self-reported

both high levels of distress themselves and, a few years later, in their children. This finding is

intriguing, but not anomalous: indeed, Pearson et al. (1994) were the first to show evidence that a

sub-set of parents (e.g., retrospectively defined earned-secures) were effective caregivers despite

32

reporting relatively high levels of depressive symptomatology. Reflecting on the Pearson et al.

(1994) and our own findings, one might speculate that secure states of mind regarding earlier

attachment experiences function as relational “buffers” or protective factors, moderating (e.g.,

attenuating) the very reliable and robust associations typically observed between maternal

depression and sub-optimal parenting.

This research owes a great deal to prior studies focused on the developing construct of

earned-security (Pearson et al., 1994; Phelps et al., 1997, Paley et al., 1999). Nevertheless, the

findings of this study do indeed clearly call into question the operational validity of retrospective

systems that require a self-reported “diagnosis” of early adversity for assessing change in the

quality of early attachments. This position in no way seeks to draw attention away from Pearson

and colleague’s (1994) earned-secure group— there is no doubt much to be learned still about

why certain autonomous participants report more harsh childhood experiences than other secure

adults in the AAI, as well as the role played by childhood and maternal depression in the early

development of these individuals. We do recommend on the basis of these data, however, that

the term “earned-secure” should be reserved for describing individuals whose encounters with

adversity can be operationalized prospectively, preferably on the basis of observational data

demonstrating positive change in insecure attachments over time. This issue is evidently

nontrivial: the analyses presented here demonstrate empirically that differing operationalizations

of earned-security can change our understanding of the nature of the construct itself, both in

terms of its childhood antecedents and consequences in adulthood.

One further caveat is in order: As is emphasized throughout this report, earned-security

was defined here retrospectively on the basis of a system articulated first by Pearson et al. (1994)

and as developed in the writings of Phelps et al. (1997) and Paley et al. (1999). Both the inferred

33

experience group means (see Table 1) as well as the findings presented herein vis-à-vis

depressive symptomatology and relational adaptation in adulthood (see Table 4) bear a striking

similarity to those described in these seminal studies, implying fidelity with the operational

definition used in prior work. In further applying this standard operationalization, we found no

evidence that earned-secures so defined experience either a higher rate of infant insecurity

compared to retrospective continuous-secures or pervasively malevolent early experiences in

their formative relationships. Nonetheless, we recognize that it is still possible that a

retrospective criterion might be developed for earned- and continuous-security that appropriately

differentiates groups in terms of their earlier childhood experiences. It should be pointed out,

however, that our presently limited data suggest that this may prove to be difficult: A reviewer of

this article was concerned that, in this study, earned-secures’ reported experiences with maternal

love fell above the midpoint on the scale (a robust finding in the earned-security literature; see

Phelps et al., 1997; Paley et al, 1999 [for female participants]). As such, in supplementary

analyses we removed individuals from the earned-secure group whose scores on the “mother

loving” scale fell above the midpoint. Even in adopting this more conservative criterion,

however, retrospectively defined earned- and continuous-security status still failed to

differentiate participants in terms of their 12- and 18-month attachment histories.

We hope this paper will be regarded as a new beginning in the study of earned-security,

with a focus on prospectively assessed positive change in representational models of attachment

from childhood to adulthood. To be sure, much remains unknown about those who rise above

harsh or ineffective parenting experiences. For example, though the results of this study indicate

that those who overcome insecure attachments in infancy do not necessarily report high levels of

depressive symptomatology in adulthood, it may be that such individuals nevertheless carry with

34

them other (as yet unspecified) liabilities as well as strengths. We look forward eagerly to future

research both to expand the boundaries of our understanding of earned-security as well as to

attempt replication of the analyses presented here (see Bahadur, 1998, however, for preliminary

confirmation). Given the paucity of long-term prospective studies of attachment security, it falls

to the few extant longitudinal studies of more normative risk cohorts to ultimately fulfill this

promise (e.g., Waters, Merrick, et al., 2000; Hamilton, 2000).

35

Acknowledgements

Special thanks are offered to Stephanie D. Madsen, Megan Sampson, Jessica

Siebenbruner, and Gloria Whaley for their coding efforts on this study. The authors also wish to

acknowledge financial support for this research from the National Institute of Mental Health to

Byron Egeland, L. Alan Sroufe, and W. Andrew Collins (MH40864-08).

Portions of this paper were presented by the first and second authors at the twelfth annual

convention of the American Psychological Society, Miami Beach, FL, June 2000.

36

Author Addresses and Affiliations

All authors of this paper are affiliated with the Institute of Child Development at the

University of Minnesota. Correspondence concerning this article should be addressed to L. Alan

Sroufe or Glenn I. Roisman, Institute of Child Development, 51 East River Road, University of

Minnesota, Minneapolis, Minnesota 55455, (612) 624-1035, FAX (612) 624-6373. E-mail may

be sent either to srouf001@tc.umn.edu or roism001@tc.umn.edu.

37

Appendix

Brief description of rating scales that compose each relationship quality/parenting measure

Maternal Support and Structure (24 months)

I. Overall Quality: Overall rating of quality of support for problem solving

II. Structuring and Limit Setting (Cleanup): Clarity and effectiveness of boundary setting

Maternal Support and Structure (42 months)

I. Hostility: Mother’s expression of anger, discounting, or rejection of child

II. Respect for Autonomy: Degree to which mother supports child’s exploration

III. Structuring and Limit Setting: Clarity and effectiveness of boundary setting

IV. Supportive Presence: Mother’s expression of positive regard and emotional support

Mother-Child Quality of Process (13 years)

I. Balance I: Degree of willingness to express individual ideas freely

II. Balance III: Ability of the parent-child dyad to meet task demands

III. Emotional Engagement: Level of emotional connectedness

IV. Positive Affect: Amount of reciprocal personal regard and pleasure

Romantic Relationship Quality of Process (20-21 years)

I. Balance I: Degree of willingness to express individual ideas freely

II. Balance II: Degree to which the relationship serves individual development

III. Conflict Resolution: Ability to resolve conflict in a way that leads to mutual

satisfaction

IV. Overall Quality: Global rating of mutual caring, emotional investment, and trust

V. Secure Base: Ability to seek or provide care in a timely, contingent manner

VI. Shared Positive Affect: Amount of reciprocal personal regard and pleasure

38

References

Achenbach, T.M. (1991a). Manual for the Child Behavior Checklist/4-18 and 1991

profile. Burlington, VT: University of Vermont Department of Psychiatry.

Achenbach, T.M. (1991b). Integrative guide for the 1991 CBCL/4-18, YSR, and TRF

profiles. Burlington, VT: University of Vermont Department of Psychiatry.

Achenbach, T.M. (1997). Manual for the Young Adult Self-Report and Young Adult

Behavior Checklist. Burlington, VT: University of Vermont Department of Psychiatry.

Aguilar, B., Christian, S.D., Collins, W.A., Cook, J., Hennighausen, K.H., Hyson, D.M.,

Levy, A.K., Meyer, S.E., Roisman, G.I., Ruh, J.L., Sesma, A.G., & Vogeler-Knopp, T. (1997).

Romantic relationship assessment observational rating scales. Unpublished coding manual,

Parent-Child Project, University of Minnesota.

Aguilar, B., Sroufe, L.A., Egeland, B., & Carlson, E. (2000). Distinguishing the early-

onset/persistent and adolescence-limited antisocial behavior types: From birth to 16 years.

Development & Psychopathology, 12, 109-132.

Ainsworth, M.D.S., Blehar, M.C., Waters, E., & Wall, S. (1978). Patterns of attachment:

A psychological study of the strange situation. New York: Lawrence Erlbaum Associates.

Ainsworth, M., & Wittig, B. (1969). Attachment and exploratory behaviors of one-year-

olds in a strange situation. In B. Foss (Ed.), Determinants of infant behavior. (Vol. 4.), New

York: Barnes & Noble.

Allen, J.P., & Hauser, S.T. (1996). Autonomy and relatedness in adolescent-family

interactions as predictors of young adults’ states of mind regarding attachment. Development &

Psychopathology, 8, 793-809.

39

Bahadur, M.A. (1998). The continuity and discontinuity of attachment: A longitudinal

study from infancy to adulthood. Unpublished doctoral dissertation, New York University.

Beckwith, L., Cohen, S.E., & Hamilton, C.E. (1999). Maternal sensitivity during infancy

and subsequent life events relate to attachment representation at early adulthood. Developmental

Psychology, 35, 693-700.

Block, J.H., & Block, J. (1980). The role of ego-control and ego-resiliency in the

organization of behavior. In W.A. Collins (Ed.), University of Minnesota Symposium on Child

Psychology. (Vol. 13, pp. 39-101). Hillsdale, NJ: Erlbaum.

Bowlby, J. (1973). Attachment and loss: Separation. (Vol. 2.), New York: Basic Books.

Bradburn, N.M. (1969). The structure of psychological well being. Chicago: Aldine.

Cohen, J. (1992). A power primer. Psychological Bulletin, 112, 155-159.

Cohen, J. (1994). The earth is round (p < .05). American Psychologist, 49, 997-1003.

Collins, W.A., & Sroufe, L.A. (1999). Capacity for intimate relationships: A

developmental construction. In W. Furman, B.B. Brown, & C. Feiring (Eds.), The development

of romantic relationships in adolescence: Cambridge studies in social and emotional

development. (pp. 125-147). New York, NY, US: Cambridge University Press.

Cohn, D.A., Silver, D.H., Cowan, C.P., Cowan, P.A., & Pearson, J.L. (1992). Working

models of childhood attachment and couple relationships. Journal of Family Issues, 13, 432-449.

Egeland, B., & Brunquell, D. (1979). An at-risk approach to the study of child abuse:

Some preliminary findings. Journal of the American Academy of Child Psychiatry, 8, 219-235.

Egeland, B., Carlson, E., & Sroufe, L.A. (1993). Resilience as process. Development &

Psychopathology, 5, 517-528.

40

Egeland, B., Jacobvitz, D., & Sroufe, L.A. (1988). Breaking the cycle of abuse. Child

Development, 59, 1080-1088.

Fox, N.A. (1995). Of the way we were: Adult memories about attachment experiences

and their role in determining infant-parent relationships: A commentary on van IJzendoorn

(1995). Psychological Bulletin, 117, 404-410.

George, C., Kaplan, N., & Main, M. (1985). Adult Attachment Interview. Unpublished

manuscript, University of California, Berkeley.

Grice, H.P. (1975). Logic and conversation. In P. Cole & J.L. Moran (Eds.). Syntax and

semantics III: Speech acts. (pp. 41-58). New York: Academic Press.

Hamilton, C.E. (2000). Continuity and discontinuity of attachment from infancy through

adolescence. Child Development, 71, 690-694.

Hays, W.L. (1988). Statistics, 4th Edition. Fort Worth, TX: Holt, Rinehart, & Winston,

Inc.

Hennighausen, K.H. (1999). Developmental antecedents of young adult romantic

relationships. Unpublished doctoral dissertation, University of Minnesota.

Henry, B., Moffit, T., Caspi, A., Lanley, J., & Silva, P. (1994). On the “remembrance of

things past”: A longitudinal evaluation of the retrospective method. Psychological Assessment,

6, 92-101.

Hesse, E. (1999). The Adult Attachment Interview: Historical and current perspectives.

In J. Cassidy & P.R. Shaver (Eds.), Handbook of attachment: Theory, research, and clinical

applications. (pp. 395-433). New York, NY, USA: The Guilford Press.

Luthar, S.S., Cicchetti, D., & Becker, B. (2000). The construct of resilience: A critical

evaluation and guidelines for future work. Child Development, 71, 543-562.

41

Main, M., & Goldwyn, R. (1998). Adult attachment scoring and classification systems,

Version 6.2. Unpublished manuscript, University of California, Berkeley.

Matas, L., Arend, R.A., & Sroufe, L.A. (1978). Continuity of adaptation in the second

year: The relationship between quality of attachment and later competence. Child Development,

49, 547-546.

Paley, B., Cox, M.J., Burchinal, M.R., & Payne, C.C. (1999). Attachment and marital

functioning: Comparison of spouses with continuous-secure, earned-secure, dismissing, and

preoccupied attachment stances. Journal of Family Psychology, 13, 580-597.

Pearson, J.L., Cohn, D.A., Cowan, P.A., & Cowan, C.P. (1994). Earned- and continuous-

security in adult attachment: Relation to depressive symptomatology and parenting style.

Development & Psychopathology, 6, 359-373.

Phelps, J.L., Belsky, J., & Crnic, K. (1997). Earned security, daily stress, and parenting:

A comparison of five alternative models. Development & Psychopathology, 10, 21-38.

Radloff, L.S. (1977). The CES-D scale. A self-report depression scale for research in the

general population. Applied Psychological Measurement, 1, 385-401.

Roisman, G.I., Madsen, S.D., Hennighausen, K.H., Sroufe, L.A., & Collins, W.A. (2001).

The coherence of dyadic behavior across parent-child and romantic relationships as mediated by

the internalized representation of experience. Attachment & Human Development, 3, 156-172.

Sroufe, J.W. (1991). Assessment of parent-adolescent relationships: Implications for

adolescent development. Journal of Family Psychology, 5, 21-45.

Sroufe, L.A., Carlson, E.A., Levy, A.K., & Egeland, B. (1999). Implications of

attachment theory for developmental psychopathology. Development & Psychopathology, 11, 1-

13.

42

Sroufe, L.A., Egeland, B., & Kreutzer, T. (1990). The fate of early experience following

developmental change: Longitudinal approaches to individual adaptation in childhood. Child

Development, 61, 1363-1373.

Sroufe, L.A., & Fleeson, J. (1986). Attachment and the construction of relationships. In

W. Hartup and Z. Rubin (Eds.), Relationships in development. Hillsdale, NJ: Erlbaum.

Sroufe, L.A., & Waters, E. (1977). Attachment as an organizational construct. Child

Development, 48, 1184-1199.

Thompson, R.A. (2000). The legacy of early attachments. Child Development, 71, 145-

152.

van IJzendoorn, M.H. (1995a). Adult attachment representations, parental

responsiveness, and infant attachment: A meta-analysis on the predictive validity of the Adult

Attachment Interview. Psychological Bulletin, 117, 387-403.

van IJzendoorn, M.H. (1995b). Of the way we are: On temperament, attachment, and the

transmission gap: A rejoinder to Fox (1995). Psychological Bulletin, 117, 411-415.

Waters, E., Hamilton, C.E., & Weinfield, N.S. (2000). The stability of attachment

security from infancy to adolescence and early adulthood: General introduction. Child

Development, 71, 678-683.

Waters, E., Merrick, S., Treboux, D., Crowell, J., & Albersheim, L. (2000). Attachment

security in infancy and early adulthood: A twenty-year longitudinal study. Child Development,

71, 684-689.

Weinfield, N.S., Sroufe, L.A., & Egeland, B. (2000). Attachment from infancy to

adulthood in a high-risk sample: Continuity, discontinuity, and their correlates. Child

Development, 71, 695-702.

43

Yarrow, M.R., Campbell, J.D., & Burton, R.V. (1970). Recollections of childhood: A

study of the retrospective method. Monographs of the Society for Research in Child

Development, 35, 1-83.

44

Footnotes

1 Note that Main and Goldwyn (1998) advocate for extremely stringent criteria in the

retrospective determination of earned- versus continuous/originally-secure status. However, as

no researchers have as yet adopted these stringent criteria, the present study uses as a point of

departure the more liberal standard developed by Pearson et al. (1994). See Methods section for

further exploration of Main and Goldwyn’s recent operationalization of earned-security.

2 Ten transcripts could not be classified, half due to equipment failures. An additional

seven U/F transcripts were dropped from analysis.

3 It should be noted that continuous-secures were rated as marginally more coherent than

earned-secures (t [44] = 2.00, p = .05), a finding with precedent in Phelps et al. (1997). As the

two secure groups were rated above the midpoint on the coherence of mind scale and had

significantly higher coherence of mind means than insecures, however, both earned- and

continuous-secures can be regarded as producing qualitatively “coherent” discourse in absolute

terms as well as relative to insecures.

4 Degrees of freedom vary in analyses regarding experience scales as in several cases

individuals were assigned “cannot rate” codes and therefore dropped from analysis. Note also

that statistics for the father rejecting and neglecting scales do not assume equal variances across

groups as Levene’s test was significant for these analyses. In all cases estimated degrees of

freedom in such analyses are rounded to the nearest whole number.

5 Given the sizable correlation between externalizing and internalizing symptoms, it

might be argued that earned-secures’ elevated distress actually reflects higher levels of

aggression relative to other security groups. In fact, post hoc analyses revealed no differences

between earned-secures and other security groups on externalizing symptoms in either

45

kindergarten/grade 1 (mother-report) or at age 16 (self- or mother-report). We thank an

anonymous reviewer for encouraging this rigorous test of symptom specificity.

46

Table 1

Retrospectively defined security group inferred experience and coherence of mind means and

standard deviations

_____________________________________________________________________________ AAI Scale Insecure Earned-Secure Continuous-Secure

_____________________________________________________________________________________________

Loving-M 4.60 (1.61) 5.46 (1.82) 6.50 (1.06)

Loving-F 3.41 (1.40) 2.56 (1.22) 5.73 (1.34)

Rejecting-M 2.00 (1.49) 2.52 (2.03) 1.36 (.62)

Rejecting-F 2.13 (1.75) 5.35 (2.61) 1.61 (1.23)

Neglecting-M 1.95 (1.62) 2.98 (2.40) 1.07 (.23)

Neglecting-F 3.00 (3.09) 6.54 (2.18) 1.82 (1.26)

Coherence of Mind 3.07 (.87) 5.71 (1.04) 6.36 (1.18)

______________________________________________________________________________

Note. Scales range from 1 (characteristic absent) to 9 (characteristic pervasive). Standard

deviations are in parentheses. M: Mother; F: Father. Experience scale means for the insecure

group are based on a random sampling of 25 cases. All other values represent the means for the

entire group.

47

Table 2

Retrospective by prospective security status

_____________________________________________________________________________ Retrospective Operationalization Earned-Secure Continuous-Secure_____________________________________________________________________________________________

Prospective Operationalization (12 mo.)

Earned-Secure 9 10

Continuous-Secure 15 11

Prospective Operationalization (18 mo.)

Earned-Secure 5 6

Continuous-Secure 16 15

______________________________________________________________________________

Note. Retrospective classification uses AAI data only (state of mind information and inferred

experience ratings), prospective classifications use cross-tabulation of infant Strange Situation

data and AAI state of mind information. κ = -.10, p = .49 for retrospective operationalization by

12-month Strange Situation prospective operationalization; κ = -.05, p = .73 for retrospective

operationalization by 18-month Strange Situation prospective operationalization.

48

Table 3

Retrospective and prospective security classifications by sex

_____________________________________________________________________________ Insecure Earned-Secure Continuous-Secure_____________________________________________________________________________________________

Retrospective Operationalization

Male 60 9 14

Female 47 15 8

Prospective Operationalization (12 mo.)

Male 60 11 12

Female 47 8 14

Prospective Operationalization (18 mo.)

Male 60 5 18

Female 47 6 13______________________________________________________________________________

Note. Χ2(1) = 3.14, p = .08 for retrospective operationalization (secure groups only). Χ2(1) = .61,

p = .44 for prospective operationalization based on 12-month Strange Situation and Χ2(1) = .52,

p = .47 for prospective operationalization based on 18-month Strange Situation (secure groups

only).

49

Table 4

Young Adult Sequelae: Self-reported internalizing distress (23 yrs.) and observed romantic

relationship quality (20-21 yrs.) means and standard deviations by retrospectively and

prospectively defined security group

_____________________________________________________________________________ Insecure Earned-Secure Continuous-Secure _____________________________________________________________________________________________

Retrospective Operationalization

YASR Internalizing Distress 48.53 (10.61) 48.91 (9.74) 44.62 (6.95) 92 22 21

Romantic Relationship Quality -.27 (.96) .43 (.80) .38 (1.21) 40 8 13

Prospective Operationalization (12 mo.)

YASR Internalizing Distress 48.53 (10.61) 46.72 (9.23) 47.50 (8.02) 92 18 24

Romantic Relationship Quality -.27 (.96) .53 (1.13) .25 (1.03) 40 7 13

Prospective Operationalization (18 mo.)

YASR Internalizing Distress 48.53 (10.61) 47.60 (9.08) 46.77 (8.58) 92 10 30

Romantic Relationship Quality -.27 (.96) 1.17 (.52) .26 (1.09) 40 2 17

_____________________________________________________________________________________________

Note. YASR data are reported using average T-values; Romantic relationship observational

group data are mean z-scores. Standard deviations are in parentheses. Group n’s appear below

means and standard deviations. YASR: Young Adult Self-Report.

50

Table 5

Antecedent Depressive Symptomatology and Parenting: Mother- and self-report of participants’

internalizing distress, mothers’ self-report of depressive symptomatology, and observations of

maternal parenting means and standard deviations by retrospectively defined security group

_____________________________________________________________________________ Insecure Earned-Secure Continuous-Secure _____________________________________________________________________________________________

Internalizing Distress (Participant)

CBCL, kindergarten/grade 1 55.94 (8.56) 60.07 (7.46) 54.61 (9.50) 101 23 22

CBCL, 16 yrs. 50.57 (11.12) 54.39 (10.25) 50.05 (7.63) 99 23 22

YSR, 16 yrs. 51.54 (9.30) 55.48 (9.83) 51.32 (7.42) 100 23 22

Depressive Symptomatology (Mother)

CES-D, 42 mo. 13.95 (10.62) 19.55 (12.01) 12.95 (9.40) 100 22 21

Antecedent Maternal Parenting

Support and Structure, 24 mo. -.11 (.95) .47 (.98) -.10 (1.19) 91 19 15

Support and Structure, 42 mo. -.09 (1.03) .17 (.70) .31 (.68) 104 20 21

Relationship Quality, 13 yrs. -.18 (1.02) .47 (.95) .12 (.89) 98 24 22

_____________________________________________________________________________

Note. CBCL and YSR data are reported using average T-values; parenting observational group

data are mean z-scores. Standard deviations are in parentheses. Group n’s appear below means

and standard deviations. CBCL: Child Behavior Checklist; YSR: Youth Self-Report; CES-D:

Center for Epidemiological Studies Depression Scale.